[BUG] batch tool is completely broken - two issues prevent any sub-tool execution

[pkunekar]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

1.29.0

Tools Package Version

0.2.22

Tools used

1. batch
2. current_time

Python Version

3.12.12

Operating System

Windows, Linux(ARM64)

Installation Method

pip

Steps to Reproduce

from strands import Agent
from strands_tools import batch, current_time

agent = Agent(tools=[batch, current_time])
result = agent("What is the current time in UTC, US/Pacific, US/Eastern, Europe/London, Asia/Tokyo, and Australia/Sydney?")

The LLM generates this tool call:

batch (invocations: [
    {'name': 'current_time', 'arguments': {'timezone': 'UTC'}},
    {'name': 'current_time', 'arguments': {'timezone': 'US/Pacific'}},
    {'name': 'current_time', 'arguments': {'timezone': 'US/Eastern'}},
    {'name': 'current_time', 'arguments': {'timezone': 'Europe/London'}},
    {'name': 'current_time', 'arguments': {'timezone': 'Asia/Tokyo'}},
    {'name': 'current_time', 'arguments': {'timezone': 'Australia/Sydney'}}
])

Expected Behavior

The batch tool should execute all 6 current_time invocations and return their results:

Batch execution completed with 6 tool(s):
✓ current_time: Success
✓ current_time: Success
...

Actual Behavior

With unpatched code (Bug 1):

The batch tool returns 0 results — invocations is silently empty:

Batch execution completed with 0 tool(s):
{"batch_summary": {"failed": 0, "successful": 0, "total_tools": 0}, "results": []}

After fixing Bug 1 only [locally] (Bug 2 surfaces):

All 6 sub-tool calls fail with ConcurrencyException:

2026-03-10 17:26:33,510 - strands.agent - INFO - Tool #1: batch
Error executing tool 'current_time': Direct tool call cannot be made while the agent is
in the middle of an invocation. Set record_direct_tool_call=False to allow direct tool
calls during agent invocation.
Error executing tool 'current_time': Direct tool call cannot be made while the agent is
in the middle of an invocation. Set record_direct_tool_call=False to allow direct tool
calls during agent invocation.
... (repeated for all 6 invocations)

Additional Context

No response

Possible Solution

Bug 1: invocations read from wrong source (line 97)

Root Cause

The bug is in src/strands_tools/batch.py line 97:

def batch(tool: ToolUse, **kwargs) -> ToolResult:
    agent = kwargs.get("agent")
    invocations = kwargs.get("invocations", [])  # ← BUG: always returns []

The batch tool is a module-based tool (uses TOOL_SPEC + function, not @tool decorator), so it is loaded as a PythonAgentTool. When PythonAgentTool.stream() calls the tool function, it passes:

# strands/tools/tools.py line 259
result = await asyncio.to_thread(self._tool_func, tool_use, **invocation_state)

• tool_use (first arg → tool param): Contains {"toolUseId": "...", "name": "batch", "input": {"invocations": [...]}}
• **invocation_state (→ **kwargs): Contains {"agent": <Agent>, "model": ..., "messages": ..., "system_prompt": ..., "tool_config": ...}

So kwargs has agent, model, messages, etc. — but NOT invocations. The invocations array is in tool["input"]["invocations"].

Compare this with DecoratedFunctionTool.stream() (used by @tool decorated functions like current_time) which correctly extracts tool_use.get("input", {}) and passes validated input to the function.

Current behavior (line 97)

invocations = kwargs.get("invocations", [])

Proposed Fix (line 97)

invocations = tool.get("input", {}).get("invocations", [])

---

Bug 2: Sub-tool calls fail with ConcurrencyException (line 112)

Root Cause

After fixing Bug 1, the batch tool correctly iterates over invocations and calls sub-tools via agent.tool.<name>(**arguments) on line 112:

result = tool_fn(**arguments)  # ← BUG: missing record_direct_tool_call=False

Since batch is called during an active agent invocation, the agent holds its invocation lock. When agent.tool.current_time(...) is called, the SDK's ToolCaller (strands/tools/_caller.py lines 88-96) tries to acquire the same lock and raises:

ConcurrencyException: Direct tool call cannot be made while the agent is in the middle
of an invocation. Set record_direct_tool_call=False to allow direct tool calls during
agent invocation.

The ToolCaller accepts record_direct_tool_call=False to skip the lock (and skip recording in message history), which is exactly what batch sub-tool calls need.

Current behavior (line 112)

result = tool_fn(**arguments)

Proposed Fix (line 112)

result = tool_fn(record_direct_tool_call=False, **arguments)

---

Summary of all changes needed in src/strands_tools/batch.py

- invocations = kwargs.get("invocations", [])
+ invocations = tool.get("input", {}).get("invocations", [])

-                     result = tool_fn(**arguments)
+                     result = tool_fn(record_direct_tool_call=False, **arguments)

Related Issues

• Bug 1 may also affect other module-based tools (TOOL_SPEC + function pattern) that read their input arguments from kwargs instead of tool["input"].
• Bug 2 would affect any tool that makes direct agent.tool.<name>() calls during an active agent invocation.

PFB Screenshot after fixing both bugs:

Related Issues

No response